Comparisons of outcomes in different burn treatment facilities are currently made using statistical procedures which attempt to adjust for differences in the severity of injuries. In other contexts it has been shown that random measurement error in a variable used to perform adjustment yields systematically biased results. Estimated percentages of body surface area burned, which are used to adjust the outcomes of different burn units, have substantial measurement error. It can be demonstrated that procedures used to compare burn units (e.g., probit analysis, logit analysis, ANCOVA, case matching) are biased. Furthermore, this bias acts in favor of units that treat less severely injured patients. Thus, the outcomes of a special burn care facility treating severely burned patients may not seem significantly better than the outcomes of a general care facility treating less severely burned patients. The proposed research will determine how large the bias is for procedures that are commonly used to compare burn outcomes. This will be accomplished by first developing a mathematical model of the data generating process. Existing data on the error in estimating burn size and depth, on the distribution of burn sizes, and on the effect of burn severity on outcome will be used to quantify the bias in those procedures. Statistical adjustment procedures which are less biased than those currently used will be developed and similarly evaluated. It is hoped that these new procedures will provide better information about the benefits of specialized burn units and a firmer knowledge base for policy decisions affecting burn care.